On Sun, 2007-03-04 at 12:08 -0800, Bruce Korb wrote: > Michael Schwendt wrote: > > There seems to be a separate "autogen-devel" package, which contains > > header files. Most likely options.h is included. > > Given that autogen is a development package anyway. GCC-from-source > builders aside, essentially anyone wanting autogen is going to want > the two files in autogen-devel. Given that and the sub-marginal > amount of space taken by the two files, I fail to understand why > it is worth the bother to separate. > Please open a bug in bugzilla. The packager of autogen needs to respond to your comments and that's the only way to be sure to get a hold of them. If you find the packager to be unreasonable other packagers can then get involved if there's a clear path forward. > >> and the two misnamed executables. > > > > The package uses the "alternatives" system for these two. Probably this is > > not obvious. /usr/bin/columns and /usr/bin/getdefs are two very generic > > file names which pollute the /usr/bin namespace and bear the risk of > > causing a conflict with other software. It would be beneficial if autogen > > called the files differently and used an own namespace, e.g. with a prefix > > or postfix. > > "getdefs" can be so treated. Go ahead and add an "ag-" prefix or something. > I don't particularly see it as a very generic name, but renaming it is > not crucial. Renaming "columns" _is_ a problem. "autogen" under various > incarnations is about a decade old with widely available public releases > available for over 6 years and it has been officially "GNU" for several > years now. "columns" is a program incorporated into many templates. > It cannot be renamed without invalidating the templates in use by at least > a few hundred people. That is not a good idea. "columns" can certainly > be treated as a "generic" program anyway. It does what its name suggests > fairly well. Nearly as well as ``ls -C'' or ``ls -x'' (does not support > different widths for different columns). By making "columns" have a > separate name, you are invalidating these templates. As stated, these programs are packaged using the alternatives system. So the end user is able to have /usr/bin/columns and /usr/bin/getdefs point at the autogen implementation or at a different implementation. So this isn't as far wrong as you make it out to be. Take a look at my system, for instance: $ rpm -ql autogen|grep bin /usr/bin/autogen /usr/bin/columns.autogen /usr/bin/getdefs.autogen /usr/bin/xml2ag $ ls -al /usr/bin/columns /usr/bin/getdefs lrwxrwxrwx 1 root root 25 Feb 25 15:39 /usr/bin/columns -> /etc/alternatives/columns lrwxrwxrwx 1 root root 25 Feb 25 15:39 /usr/bin/getdefs -> /etc/alternatives/getdefs $ ls -al /etc/alternatives/columns /etc/alternatives/getdefs lrwxrwxrwx 1 root root 24 Feb 25 15:39 /etc/alternatives/columns -> /usr/bin/columns.autogen lrwxrwxrwx 1 root root 24 Feb 25 15:39 /etc/alternatives/getdefs -> /usr/bin/getdefs.autogen Running rpm -q --scripts autogen will show you the scriptlets that are registering columns and getdefs with the alternatives system to automatically create /usr/bin/columns and /usr/bin/getdefs on installation of the package. Is alternatives the right solution to this problem? Probably not. Alternaties was invented to address things like sendmail's sendmail program vs postfix's sendmail program. Both those programs take the same options on the commandline and yield similar results through different backends. Is autogen's version of /usr/bin/columns part of POSIX, the Single Unix Specification, or LSB? If so, we have a solid foundation and alternatives (or removing alternatives as there is no alternative implementation in the system currently) makes sense. If not, then we can only hope that no other open source software chooses the generic /usr/bin/columns name for their program otherwise we all lose. -Toshio
Attachment:
signature.asc
Description: This is a digitally signed message part
-- fedora-extras-list mailing list fedora-extras-list@xxxxxxxxxx https://www.redhat.com/mailman/listinfo/fedora-extras-list
